As AI generates more code than humans can review, the validation bottleneck emerges. The solution is providing agents with dedicated, sandboxed environments to run tests and verify functionality before a human sees the code, shifting review from process to outcome.

Cursor discovered that agents need more than just code access. Providing a full VM environment—a "brain in a box" where they can see pixels, run code, and use dev tools like a human—was the step-change needed to tackle entire features, not just minor edits.

To address security concerns, powerful AI agents should be provisioned like new human employees. This means running them in a sandboxed environment on a separate machine, with their own dedicated accounts, API keys, and access tokens, rather than on a personal computer.

AI agents present a UX problem: either grant risky, sweeping permissions or suffer "approval fatigue" by confirming every action. Sandboxing creates a middle ground. The agent can operate autonomously within a secure environment, making it powerful without being dangerous to the host system.

Instead of giving an LLM hundreds of specific tools, a more scalable "cyborg" approach is to provide one tool: a sandboxed code execution environment. The LLM writes code against a company's SDK, which is more context-efficient, faster, and more flexible than multiple API round-trips.

Simply giving an AI agent thousands of tools is counterproductive. The real value lies in an 'agentic tool execution layer' that provides just-in-time discovery and managed execution to prevent the agent from getting overwhelmed by its options.

"Code Mode" is not an alternative to MCP but a more efficient way to use it. Instead of multiple sequential tool calls, the model generates a single script that executes multiple actions in a sandbox. MCP still provides the core benefits of authentication, discoverability, and a standardized, LLM-friendly API.

The simple "tool calling in a loop" model for agents is deceptive. Without managing context, token-heavy tool calls quickly accumulate, leading to high costs ($1-2 per run), hitting context limits, and performance degradation known as "context rot."

The true capability of AI agents comes not just from the language model, but from having a full computing environment at their disposal. Vercel's internal data agent, D0, succeeds because it can write and run Python code, query Snowflake, and search the web within a sandbox environment.